In 2021, the National Archives of Estonia engaged Digital Transitions’ Service division, Pixel Acuity, to build an Artificial Intelligence (AI) tool to analyze part of its historic record. The objective was to use this tool to enhance their collection with descriptive metadata that identified persons of interest in a collection of over 8,000 photographic glass plate negatives, a task that would ordinarily take years of human labor. In this presentation, we discuss our approach to accurately detecting and identifying human subjects in transmissive media, our initial findings using commercially available AI models, and the subsequent refinements made to our workflow to generate the most accurate metadata. In addition to working with commercially available AI models, we developed strategies for validation of AI-generated results without additional human supervision, and explored the benefits of building bespoke, heritage-specific AI models. By combining all of these tools, we developed a highly customized solution that greatly expedited accurate metadata generation with minimal human oversight, operated efficiently on large collections, and supported discovery of novel content within the archive.